An inverse ocean box modeling approach is used to address the question of what may have caused decreased atmospheric CO2 concentration during glacial periods. The inverse procedure seeks solutions that are consistent, within prescribed uncertainties, with both available paleo observations and box model conservation equations while relaxing traditional assumptions such as exact steady state and precise prescription of uncertain model parameters. Decreased ventilation of Southern Ocean deep water, decreased Southern Ocean air-sea gas exchange, and enhanced high-latitude biological pumping are all shown to be individually capable of explaining available paleodata constraints provided that significant calcium carbonate compensation is allowed. None of the scenarios require more than a very minor (order 1 ¿C) glacial reduction in low and midlatitude sea surface temperature although scenarios with larger changes are equally plausible. One explanation for the fairly wide range of plausible solutions is that most paleodata directly constrain the inventory of paleotracers but only indirectly constrain their fluxes. Because the various scenarios that have been proposed to explain pCO2 levels during the last glacial maximum are distinguished primarily by different fluxes, the data do not allow one to confidently chose between them. ¿ 2001 American Geophysical Union